solanabr/startup-builder-skill/incident-comms

Incident comms

The stack can detect an incident and diagnose it. This skill decides what users are told, where, and by when. It is deliberately thin on top of three packs, and links them instead of restating them:

Context handoff

At start, read .claude/context/incidents.md if present: it holds this team's ladder, roles, approved templates and open incidents. Read .claude/context/launch.md if present; an incident during a launch's L+ window also changes that runbook. Take product name and the words to avoid from .claude/context/positioning.md if present.

Writes .claude/context/incidents.md in the format at the end of severity.md, including the ## Known issues section.

Mode 1: prepare (peacetime; do this first)

1. Fill the ladder and roles in severity.md with this team's

channels, deadlines and two names per role.

1. Adapt the templates in updates.md to the product: real

action names, the vault and program accounts users can check, the support link. Get the SEV-1 versions, and any wording about user losses, approved once by counsel, and record who approved them and when.

1. Build the status page from status-page.html and deploy it

on a host and domain that do not fail with the app. Confirm someone other than the app's CI can update it.

1. List the integrators who route users to you (aggregators, wallets, partners) with a

direct contact for each. In a SEV-1 they are the fastest way to stop new users arriving.

1. Write .claude/context/incidents.md.

Mode 2: during an incident

1. Set severity from onchain reality using the ladder. Ask: can a user lose funds, act

on their funds, or only see the wrong thing? Open INC-<YYYYMMDD>-<n> in the log.

1. Acknowledge before you understand. Post update 1 from

updates.md to the status page, then link it in each channel the ladder names, within the first-update deadline. For a malicious frontend or DNS, the first line is "don't use {{DOMAIN}}", and nothing delays it.

1. Write the known-issue entry in the same minute, per

known-issue.md, so community-moderation's dedupe step links members to the status page instead of opening duplicate tickets.

1. SEV-1: switch to the [exploit track](references/exploit-track.md). The warning goes

out now; amounts, cause, compensation, messages to the attacker and any contact with regulators, law enforcement or issuers wait for counsel via crypto-legal-skill. Do not draft those here, and do not guess at them.

1. Progress updates on the ladder's cadence, even when nothing changed. Re-grade

severity on evidence, in either direction, and say so in the next update.

1. Resolve with update 3. A SEV-1 stays "Mitigated" while the cause is unknown or funds

are still held by an attacker.

Mode 3: after

• Write the user-facing account from postmortem.html: what

users lost, what is being done about it, what changes. It links the auditor's technical report and does not replace it. For a SEV-1, or any postmortem that mentions user losses, counsel sees it before it is published.

• Add the postmortem link to the status page entry and the known-issue entry, and set both

to resolved.

• If preparation failed (a late first update, an unreachable signer, a status page that went

down with the app), fix the plan in incidents.md now.

Rules

• "Funds are safe" is a statement about chain state. Say it only after the accounts have

been checked onchain.

• Every onchain action you take (pause, upgrade, authority change) is announced with its

transaction link, before someone else finds it.

• Legal, regulatory and compensation questions go to crypto-legal-skill. Say so and stop.